Machine Learning Models for Bot Detection: Building Improvements to CAPTCHAS; Exploring Political Strength of Cyberattacks 6 views

Artificial intelligence (AI) improvements have brought many interesting innovations and applications to help better our lives, but this technology has also been used to improve malicious technology. The overall encompassing problem is that improvements to AI can be used to help perform cyberattacks, so we need to research methods to defend against these cyberattacks. For example, current bots are now able to bypass modern robot detection strategies with the help of AI, making malicious botting activities more of a threat and requiring more prevention and mitigation efforts. Compared to before, malicious botting is capable of more disruption than before with drastically increased ability and performance thanks to the implementation of AI. Combined with modern development of society potentially giving more motivation and higher risk for cyberattacks, this could cause a resurgence in malicious botting that can cause widespread disruption to millions of people’s daily lives. We must research new defense technology as well as introduce new social norms regarding the internet in order to defend against future cyberattacks. Understanding potential risks factors that increase the strength of cyberattacks combined with improved security capabilities to defend against intelligent bots will ensure the stability of core online infrastructure and maintain order in society and mitigate the effects of this increased risk of botting. One avenue to mitigate the effects of cyberattacks strengthened by AI is to create new technology to defend against bots with AI capabilities. Bots with malicious intentions consist of around 37% of all internet traffic. I proposed a modernized test inspired by a Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) to prevent bots from accessing certain websites. In this updated version of CAPTCHAs, I proposed a machine learning model that trains using bots that can pass CAPTCHAs. The model’s training method was inspired by generative adversarial networks (GAN). The idea is to have two models: the generator, and the classifier. The generator attempts to create bots that try to trick our classifier, and our classifier aims to correctly identify each bot. While training itself to classify bots, it also trains these bots to improve their ability to trick our classifier, creating cycles of smarter models to simulate improved intelligence of future bots. We continuously train these models until sufficient accuracy is reached on the training bots. After the training, we expect a test accuracy of above 90% in classifying bots from human inputs. If practical, this model could be implemented to prevent bots from visiting certain websites, limiting the amount of malicious activity that can be done through bots. Another method to mitigate the effects of cyberattacks is to understand where they get their political strength from and minimize the presence of these factors in society. I had the research question of what do the design of politically motivated cyberattacks tell us about the vulnerable points of the internet, and how this contributes to the political strength of cyberattacks. I claimed that reliance on the internet for important societal functions, the trend of moving services online, and the lack of emphasis on punishing cyberattacks lead to a strengthened political effect of cyberattacks. I used cyberattack cases from the Russia-Ukraine War to provide evidence that these attacks were designed to take advantage of these features of society to perform cyberattacks with great political impact. Ultimately, I concluded that there needs to be caution when moving core societal components online, and there should be expected failure and a backup plan in case. I also found that there needs to be a different view of cyberattacks that what society currently has. Cyberattacks need to be treated more seriously so that deterrents can be in place to prevent the attractiveness of cyberattacks as a way of political gain.

BS (Bachelor of Science)

Cyberattacks; Bot Detection

School of Engineering and Applied Science Bachelor of Science in Computer Science Technical Advisor: Briana Morrison STS Advisor: Kent Wayland

All rights reserved by the author (no additional license for public reuse)

2026-05-07